When an experimentally determined protein structure is not available, homology or comparative modeling can frequently predict a useful model of a given sequence by relying on its similarity to proteins of known three-dimensional structure. The broad goal of the proposal is to increase the usefulness of comparative modeling for functional annotation of proteins. This goal will be achieved by automated methods that will minimize and detect errors in comparative models, and also be applicable to sequences only remotely related to known protein structures. The specific aims are: 1. To increase the sensitivity of detecting known protein structures related to the modeled sequence, by identifying intermediate sequences matching both the known structure and the modeled sequence. 2. To minimize errors in protein sequence-structure alignments, by using an iterative process of alignment, model building, and model assessment, as well as structure and multiple sequence information. 3. To minimize model errors in sidechain packing, distortions and shifts in the core regions, and errors in the modeling of loops. This aim will be achieved by relying on satisfaction of spatial restraints from template structures and energy terms from molecular mechanics, including an implicit solvation model. 4. To improve the assessment of the overall and local errors in a given comparative model, by relying on multivariate model assessment functions. 5. To automate and implement all the methods in the protein structure modeling program MODELLER. 6. To facilitate structure determination of the complete Saccharomyces cerevisiae ribosome, by providing comparative models for the fitting of proteins into the cryo-electron microscopy map of the ribosomal particle. Achieving these aims will augment other projects in the laboratory, including modeling of protein sequences on a genomic scale, a comprehensive database of models for all sequences detectably related to at least one known structure, and a web server for comparative protein structure modeling. More broadly, the specific aims are especially timely because of the advent of structural and functional genomics. The improvements will leverage the genomics effort by significantly increasing both the number of modeled proteins and the model accuracy.